Reaction Graph Networks for Inorganic Synthesis Condition Prediction

The integration of advanced machine learning (ML) techniques with density functional theory (DFT) has significantly enhanced the prediction of stable material structures. However, translating these computational predictions into successful laboratory syntheses—whether by autonomous labs or human scientists—remains a challenge due to the complex optimization of solid-state reaction parameters. In this work, we model inorganic <i>solid-state</i> reactions through a graph neural network to incorporate precursor interactions into the prediction of synthesis conditions. We evaluated our methodology on literature-reported syntheses of clean energy materials, demonstrating enhanced performance over prior synthesis condition predictions. Our approach has the potential to streamline the transition from computational predictions to actual material synthesis, potentially acclerating the discovery of new materials.